Transcriptome-wide mega-analyses reveal joint dysregulation of immunologic genes and transcription regulators in brain and blood in schizophrenia

Abstract

The application of microarray technology in schizophrenia research was heralded as paradigm-shifting, as it allowed for high-throughput assessment of cell and tissue function. This technology was widely adopted, initially in studies of postmortem brain tissue, and later in studies of peripheral blood. The collective body of schizophrenia microarray literature contains apparent inconsistencies between studies, with failures to replicate top hits, in part due to small sample sizes, cohort-specific effects, differences in array types, and other confounders. In an attempt to summarize existing studies of schizophrenia cases and non-related comparison subjects, we performed two mega-analyses of a combined set of microarray data from postmortem prefrontal cortices (n=315) and from ex-vivo blood tissues (n=578). We adjusted regression models per gene to remove non-significant covariates, providing best-estimates of transcripts dysregulated in schizophrenia. We also examined dysregulation of functionally related gene sets and gene co-expression modules, and assessed enrichment of cell types and genetic risk factors. The identities of the most significantly dysregulated genes were largely distinct for each tissue, but the findings indicated common emergent biological functions (e.g. immunity) and regulatory factors (e.g., predicted targets of transcription factors and miRNA species across tissues). Our network-based analyses converged upon similar patterns of heightened innate immune gene expression in both brain and blood in schizophrenia. We also constructed generalizable machine-learning classifiers using the blood-based microarray data. Our study provides an informative atlas for future pathophysiologic and biomarker studies of schizophrenia.

Keywords: Blood; Brain; Gene expression; Innate immunity; Random forests; Schizophrenia; Support vector machine; Transcriptome.

Figure 1. Cross-Tissue Comparison of Significantly Dysregulated Gene Sets (Bonferroni p <0.05)

The results of the permutation-based gene-set analyses are shown (Panel A), highlighting gene sets that were significantly dysregulated in both tissues. A total of 745 gene sets were dysregulated (out of 9254 tested) based on single-gene test statistics from the brain mega-analysis. A total of 526 gene sets were dysregulated (out of 9256 tested) based on single-gene test statistics from the blood mega-analysis. For the purpose of cross-tissue comparison, gene sets with either an absolute effect (i.e., all genes in the target set) or a mixed effect (i.e., a subset of the genes in the target set) were considered to be directionally dysregulated. Among the dysregulated gene sets, 263 were common to both tissues (Bonferroni-corrected hypergeometric p < 1.7×10⁻¹⁵⁸, and among these, 255 showed evidence of up-regulation across tissues (p < 2.8×10⁻¹⁹⁸), while only 4 showed evidence of down-regulation across tissues (p < 4.6×10⁻⁴). Detailed methods for gene set analysis and the reference databases can be found in the Supplementary Methods. Among dysregulated gene sets corresponding to the Molecular Signature Database’s (Broad Institute) Hallmark category, we assessed whether SZ cases showed significant evidence for heterogeneity using a previously developed approach described in the Supplementary Methods. Within the brain data, we observed significant 2-group clustering of SZ cases based on the expression values corresponding to 5 gene sets showing a main-effect of upregulation in SZ (Panel B); cases are shown such that the individuals belonging to the cluster driving the up-regulation effect are depicted in red. These results suggest that different SZ cases contribute to the observed dysregulation in distinct biological pathways. Within the blood data, we observed significant 2-group clustering for a single gene set which showed a main-effect of down-regulation among SZ cases (Panel C).

Figure 2. Cross-Tissue Comparison of Significantly Dysregulated Genes (FDR q <0.05)

(A) Based on the mega-analyses, we identified the most significantly dysregulated genes in brain (n = 92) and blood (n = 2238) at a relatively conservative threshold (FDR q < .10). A total of 10 genes were common to both lists; 7 genes were coordinately up-regulated and 1 gene was coordinately down-regulated across tissues. The degree of cross-tissue overlap for each of the displayed intersections was non-significant based on hypergeometric test statistics.

Figure 3. Co-expression Modules Nominally Associated with SZ in Brain (p < 0.05)

Comparison of module eigengene expression values (unadjusted for covariates) between SZ cases and unaffected comparisons within the “green” and “salmon” co-expression modules identified by the WGCNA R package (A and E, respectively), which were nominally associated with SZ from linear mixed model (uncorrected p < 0.05). We cross-referenced the set of dysregulated genes identified in the brain mega-analysis (q < .1) with the top 25 genes in each module ranked by intramodular connectivity (overlaps denoted by asterisk *) (B and F). In panels B and F, the top 5 “hub” genes are found in the innermost circle. Modules were biologically characterized by testing for enrichment of brain cell-type signatures (C and G) and annotations by a pathway-based approach (D and H). In panels D and H, annotations that surpassed a BH p < 0.05 (cutoff depicted by vertical dotted line) from hypergeometric tests are shown (represented as −log10[P]).

Figure 4. A Co-expression Module that was Significantly Associated with SZ in Blood (BH p < 0.05)

Comparison of module eigengene expression values (unadjusted for covariates) between SZ cases and unaffected comparisons within the “darkolivegreen” co-expression module identified by the WGCNA R package (A), which was significantly associated with SZ based on linear mixed model test (BH p < 4.4×10⁻⁶). We cross-referenced the set of dysregulated genes identified in the blood mega-analysis (q < .1) with the top 25 genes ranked by intramodular connectivity in this module (overlaps denoted by asterisk *) (B). In panel B, the top 5 “hub” genes are found in the innermost circle. To biological characterize this module, a pathway-based approach was used to test for significant enrichment biological annotations mapping to “darkolivegreen” genes (C). In panel C, annotations that surpassed a BH p < 0.05 (cutoff depicted by vertical dotted line) from hypergeometric tests are shown (represented as −log10[P]).